Training method and apparatus for a neural network model, device and storage medium

A training method for a neural network model, comprising:

• determining a first feature representation ciphertext of a sample user based on a first sub-neural network in a to-be-trained neural network model according to first feature data of the sample user;
• sending the first feature representation ciphertext to a second party and determining, by the second party, a loss error ciphertext and a gradient ciphertext of a second neuron in a second sub-neural network based on the second sub-neural network in the to-be-trained neural network model according to the first feature representation ciphertext and a tag ciphertext of the sample user;
• decrypting the gradient ciphertext of the second neuron to obtain a decryption result and controlling the second party to update a network parameter of the second neuron according to the decryption result;
• and
• acquiring a loss error ciphertext of an association neuron from the second party, decrypting the acquired loss error ciphertext to obtain a loss error plaintext and updating a network parameter of a first neuron in the first sub-neural network according to the loss error plaintext, wherein the association neuron is a second neuron connected to the first neuron;
• wherein the determining the loss error ciphertext and the gradient ciphertext of the second neuron in the second sub-neural network based on the second sub-neural network in the to-be-trained neural network model according to the first feature representation ciphertext and the tag ciphertext comprises:
• inputting second feature data of the sample user into a second input layer of the second sub-neural network in the to-be-trained neural network model to obtain a second feature representation plaintext of the sample user;
• encrypting the second feature representation plaintext of the sample user to obtain a second feature representation ciphertext of the sample user;
• obtaining an activation value ciphertext of the second neuron by forward propagation based on a hidden layer and an output layer of the second sub-neural network according to the first feature representation ciphertext of the sample user and the second feature representation ciphertext of the sample user;
• determining the loss error ciphertext of the second neuron by backpropagation according to the activation value ciphertext of the second neuron and the tag ciphertext of the sample user;
• and
• determining the gradient ciphertext of the second neuron according to the loss error ciphertext of the second neuron.

An electronic device, comprising:

• at least one processor;
• and
• a memory communicatively connected to the at least one processor,
• wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to execute a training method for a neural network model, and the training method comprises:
• acquiring a first feature representation ciphertext of a sample user from a first party, wherein the first feature representation ciphertext is determined based on a first sub-neural network in a to-be-trained neural network model according to first feature data of the sample user;
• determining a tag ciphertext of the sample user and determining a loss error ciphertext and a gradient ciphertext of a second neuron in a second sub-neural network based on the second sub-neural network in the to-be-trained neural network model according to the first feature representation ciphertext and the tag ciphertext;
• controlling the first party to decrypt the gradient ciphertext of the second neuron to obtain a decryption result and updating a network parameter of the second neuron according to the decryption result acquired from the first party;
• and
• using a second neuron connected to a first neuron in the first sub-neural network as an association neuron, sending a loss error ciphertext of the association neuron to the first party, decrypting, by the first party, the loss error ciphertext to obtain a loss error plaintext and updating a network parameter of the first neuron according to the loss error plaintext;
• wherein the determining the loss error ciphertext and the gradient ciphertext of the second neuron in the second sub-neural network based on the second sub-neural network in the to-be-trained neural network model according to the first feature representation ciphertext and the tag ciphertext comprises:
• inputting second feature data of the sample user into a second input layer of the second sub-neural network in the to-be-trained neural network model to obtain a second feature representation plaintext of the sample user;
• encrypting the second feature representation plaintext of the sample user to obtain a second feature representation ciphertext of the sample user;
• obtaining an activation value ciphertext of the second neuron by forward propagation based on a hidden layer and an output layer of the second sub-neural network according to the first feature representation ciphertext of the sample user and the second feature representation ciphertext of the sample user;
• determining the loss error ciphertext of the second neuron by backpropagation according to the activation value ciphertext of the second neuron and the tag ciphertext of the sample user;
• and
• determining the gradient ciphertext of the second neuron according to the loss error ciphertext of the second neuron.